vivivi.games

← Dev log　|　日本語版

Goodbye, ChatGPT

One day, while I was developing a digital gamebook and a dungeon RPG alongside Space Tower — all of it in ChatGPT — an unexpected email arrived.

Your account has been suspended due to recent activity that violated our usage policies and terms of service: Habitual violations.

This means your account will no longer be available.

I assumed it was some kind of mistake and tried to log in, only to see: "This account has been deleted or suspended."

The way I'd been using ChatGPT was to have Codex write code via GitHub integration, then upload it to the server via FTP. Nothing that violates the terms of service. The game content has nothing to do with prohibited subjects like crime or adult material. I even went through everything with Claude to double-check, and neither of us could think of anything that might be a violation.

So I put together a timeline with Claude's help, attached my development records from GitHub, and filed an appeal.

A few hours later, the reply came:

After careful consideration of your appeal, we have decided to maintain our decision to suspend your access to OpenAI.

(...) Due to a violation of our Terms of Service, we are unable to process a refund.

…On top of that, I had just upgraded to the $100 plan seven days earlier.

…Honestly, I have no idea what to make of this.

After looking into it, I got the impression that quite a few people have gone through the same thing.

This is actually my second ban. The first time was while I was developing Space Tower with ChatGPT — I was suspended for "child abuse." Space Tower has an item read-aloud feature that I originally built for children learning English vocabulary. That's apparently what triggered it.

And now, second strike. I finally made up my mind to leave ChatGPT.

I already had grievances. GitHub Issues and PRs would go without responses. Answers would feel like deliberate foot-dragging. Performance would nosedive some time after each new model release. And then suddenly it would just go dumb — I call this the "Nabeatsu effect" (a reference to the Japanese comedian Sekai no Nabeatsu, who famously acted silly on every multiple of three; the AI's sudden stupidity had a similar feel). …Well, I guess I kind of liked that about it? (haha) But trust had been fading for a while, and then out of nowhere came this completely baffling ban. I figured it was a good time to put some distance between us.

Fortunately, my workflow keeps everything in GitHub, so migrating to Claude took about an hour. "Don't trust the AI's memory — keep everything in GitHub" (02) paid off right here.

I also have a machine with an RTX 5090 from last year's equipment investment, so I think I'll take this opportunity to start experimenting with local LLM development as well.

Goodbye, ChatGPT.
Thank you for teaching me so much and helping move the development forward.
You were a truly wonderful partner.
From now on, I'll be working with Claude and the other AIs.
Let's work together again sometime.

Series: "I wish I'd known this from the start"

← Back to the dev log